Gain Improvement of a DRA Using Deep Reinforcement Learning with Polygon Mesh Deformation

A new approach for improving the gain of a hemi-spherical dielectric resonator antenna (DRA) operating at its fundamental mode using Deep Reinforcement Learning (RDL) is introduced. Unlike conventional optimization method, the key idea here is to avoid antenna parametrization during the optimization process so that the surface shape of the DRA can be directly manipulated. The simulated results show that the proposed DRA structure can achieve a high gain over the bandwidth of 2.23-2.7 GHz with its maximum value of 7.4 dBi. This is higher than the typical value dictated by this DRA shape and the data available in the literature.